I design and build AI systems for regulated businesses.
Strategy, architecture, evaluation and human–machine research. Built first in regulated finance, then carried into any place a system has to be trusted.
AI systems
A dependency-ordered route from foundations to production.
ResearchMind and intelligence
Cognition, consciousness, and governed human–machine architectures.
Banking systems
Detailed, anonymised designsGoverned Agent Certification Fabric
An enterprise portfolio control plane that versions every agent's identity, reachable authority, dependencies, evaluation evidence and recovery obligations before it can enter service.
Open system architecture → Architecture caseNetwork-Aware Screening Investigation
An enterprise financial-crime workflow that resolves ambiguous parties and ownership paths, presents competing identity hypotheses and leaves every disposition with an authorised investigator.
Open system architecture → Architecture casePurpose-Bound Context Compiler
An enterprise shared context service that admits only authorised, applicable and fresh evidence for a decision, then preserves enough lineage to reconstruct what the system knew.
Open system architecture →Recent writing
Where should autonomy stop? What evidence proves an agent behaved safely? What context can a regulated institution stand behind at the decision point?
Deterministic-First Agents: The Certainty Gradient for Governed Banking Automation
A production architecture for decomposing banking workflows into deterministic controls, calibrated models and bounded reasoning, so autonomy is earned by evidence rather than granted by default.
Selected analysisModel-Risk Evidence for Agentic Systems: From Outcome Accuracy to Trajectory Assurance
A model-risk and production-assurance architecture for evaluating an agent's complete decision path: evidence, tools, order, policy, recovery, approvals and observed effects, not only its final answer.
Selected analysisGoverned Context Architecture: Reconstruction, Provenance and Active Memory for Long-Horizon Agents
A bank-grade context architecture that reconstructs current state, enforces authorization before retrieval, links claims to evidence and treats memory writes as governed actions rather than unlimited conversation history.